Prognostic Analysis of Synchronous Generator Winding Failures using Artificial Neural Network

Abstract

Power generation systems, which play significant role in today's world, operate on synchronous generators that are reliable and well-performing. This research focuses on the prognostic analysis of synchronous generator winding failures using artificial neural networks (ANNs). The research begins with understanding what synchronous generators are, types of faults that occur in them, and techniques used for diagnosing them. Our research work starts with collection and preprocessing of fault and healthy datasets, including analysis of parameters such as current phases, phase-to-phase voltages, voltage phase-to-neutral, winding temperature, and neutral current. We then developed a Multi-Layer Perceptron (MLP) neural network to classify fault conditions and evaluated it using metrics, including precision, recall, F1-score, accuracy, confusion matrix, and ROC-AUC score, to ensure a comprehensive understanding of its performance and validation.

Seeing the results, it was concluded that the proposed ANN-based approach effectively identifies winding failures in synchronous generators, with high accuracy and an AUC score of 0.98. This suggest that integrating this model with real-time condition monitoring systems can improve fault detection, reduce downtime, and enhance reliability. In future, the researchers can focus on refining the dataset, optimizing neural network, and integrating the model with explainable AI. This study paves the way for predictive maintenance framework for efficiency of power generation systems.